Upgrade to Pro — share decks privately, control downloads, hide ads and more …

Object Recognition with Grassmannian Manifolds

F51da3602a201d3dca41fa924e29b29d?s=47 Will Speak
May 09, 2013
46

Object Recognition with Grassmannian Manifolds

Final year project presentation for my BEng.

The abstract of the accompanying project report:
"The use of Grassmannian Manifolds has proved a novel solution to the problem of facial recognition. In this project their application to the more general field of object recognition is assessed, a discussion of the existing literature is undertaken, and an example implementation is created. A comparison is made of three different measures of subspace distance: Geodesic, Projection and Binet-Cauchy. A discussion of the process of implementing the algorithm is made. The effectiveness of the implementation is evaluated using the ETH-80 data set. The results are interpreted and the effectiveness of each of the distance measures is assessed."

F51da3602a201d3dca41fa924e29b29d?s=128

Will Speak

May 09, 2013
Tweet

Transcript

  1. OBJECT RECOGNITION WITH GRASSMANNIAN MANIFOLDS Will Speak (wgs502)

  2. OBJECTIVES Implement Grassmannians

  3. LITERATURE REVIEW Object recognition is complex Many steps involved Level

    of vision Sampling the scene Identifying objects
  4. LITERATURE REVIEW x. Lui et. al. 2004 LDA Graph-Embedding LDA

    Kernel LDA
  5. M. Harandi, C. Sanderson, S. Shirazi, and B. Lovell GRAPH

    EMBEDDING DISCRIMINANT ANALYSIS ON GRASSMANNIAN MANIFOLDS FOR IMPROVED IMAGE SET MATCHING
  6. GRASSMANN KERNEL MAPPING [M. Harandi et. al.]

  7. T. Wang and P. Shi KERNEL GRASSMANNIAN DISTANCES AND DISCRIMINANT

    ANALYSIS FOR FACE RECOGNITION FROM IMAGE SETS
  8. IMPLEMENTATION Python programming language MDP • NumPy • PIL

  9. REPRESENTING A MANIFOLD Load images from disk Vectorise images PCA

    on image set Represents a manifold span
  10. MANIFOLD DISTANCE Principle Angles Relate to geodesic distance Calculated with

    SVD Projection • Geodesic • Binet-Cauchy
  11. RECOGNITION PERFORMANCE

  12. RECOGNITION PERFORMANCE 1" 2" 3" 4" 5" 7" Projec.on" 0.863"

    0.863" 0.900" 0.913" 0.913" 0.913" Binet8Cauchy" 0.825" 0.825" 0.825" 0.838" 0.850" 0.850" Geodesic" 0.875" 0.875" 0.875" 0.888" 0.888" 0.888" 0.800" 0.810" 0.820" 0.830" 0.840" 0.850" 0.860" 0.870" 0.880" 0.890" 0.900" 0.910" 0.920" 0.930" 0.940" 0.950" Recogni(on)Rate)
  13. DEMONSTRATION

  14. THANK YOU